Anyone who thinks pollsters guilty of 'herding' ought to follow Paddy Ashdown and eat their hat. I cannot recall a time since ComRes was founded in 2003 when there has been such divergence of poll leads.
Shortly after Theresa May announced the election, on 18 April, ComRes published a poll which gave the Conservatives a whopping 25-point lead and, at 50%, gave the Tories the largest vote share of any party by any pollster since January 1991.
That lead was not sustainable, especially in the wake of the Tory Manifesto fiasco, but the range of poll leads published over the past week is huge. It currently ranges from five to 14 points. Tory vote shares range from 42% to 47% and Labour from 32% to 38%. Whether this range narrows between now and polling day remains to be seen.
There are several reasons why at ComRes we have taken a particular approach to our modelling methods. In view of the frequent questions we are asked about the reasons behind this range, we felt it would be helpful to outline those reasons.
First, turnout. If you ask different demographic groups how likely they are to vote, the answers you get translate into behaviour in different but predictable ways. In other words, when one person says that they are a ten-out-of-ten, rock-solid certainty to vote, that does not necessarily mean the same as when another person of a different age or income group says it. The voter turnout model we developed in the wake of the 2015 General Election, when we were close (but not close enough) to the final result, adjusts for that known, observable discrepancy. We are in a minority among pollsters for doing so.
This has a direct impact on our published voting intention results. While our turnout model takes account of factors other than age, broadly speaking the Conservatives have a commanding lead among those aged over 45, while Labour is ahead among younger groups. At present, more than 60% of 18-24s say they are 'absolutely certain' to vote, yet only 44% did so in 2015. So you can see how easy it might be to exaggerate Labour's vote share by overstating the propensity of young people to vote - if younger people behave in the same way they did in 2015 and indeed in previous elections. But the last time more than 60% of 18-24s voted was in 1992 - and more than 40% of them in that election voted Conservative.
Second, leadership. Although the average error in our final 2015 poll was just 1.3%, we (along with everyone else) overstated Labour and understated the Tory vote shares. We would have been spot-on had we made one further adjustment, based on how we treated people who said they did not know how they would be voting. Traditionally we have reallocated them based on which party they most identify with. But when we reallocated them by which party leader they thought in 2015 would make the best Prime Minister, our vote shares would have been less than one percentage point away from the actual result.
How would that affect today's polling? The 12-point Tory lead in our poll for the Sunday Mirror / Independent extends to 15 points using this modest methodological change.
This is potentially significant: in an election fought along presidential lines, leader perceptions matter more. We also know that one in three current Labour voters intends to support that party despite misgivings over Jeremy Corbyn - so when undecideds consider their decision on polling day, it is quite possible that the party with which a voter most identifies will be less important than how they rate different leaders.
What does all this mean for the current campaign? Very simply, if voters behave in the way they broadly did in 2015, then the Conservatives remain on track for a 100-plus majority. This seems, on present assumptions, the most likely outcome. Older people appear more motivated than younger people to vote, most of UKIP's 2015 vote is going to the Conservatives (and that Party is not even standing in around half of all constituencies), May beats Corbyn on most 'best for' measures, and Labour's core vote lacks motivation.
However, with just nine days to go, this remains a unique election - as quiet as 2005, yet highly volatile. Even if Mrs May continues her tenure at No 10, and if the Conservatives are in the mid-40s on polling day and Labour in the low-30s - as seems likely, translating that accurately into seat numbers is a perilous task.
Andrew Hawkins is the chairman of ComRes
This blog first appeared on the ComRes site, and can be read here